Transcript Machine Intelligence

EMBODIED INTELLIGENCE
Satpreet Arora (07d05003)
Rachit Gupta (07d05008)
Devendra Shelar (07d05010)
Amrose Birani (07005003)
Father of Embodied Intelligence
 Rodney Brooks suggested the
design of intelligent machines
through interaction with the
environment driven by
perception and action, rather
than by a pre-specified
algorithm.
 Brooks showed
that robots could be more
effective if they 'thought'
(planned or processed)
and perceived as little as
possible.
Precursors of Embodied Intelligence

In the early stages of robotics robots
were built on cybernetic principles.

Brooks proposed that vision and
locomotion are the 2 primary needs
of natural intelligence.
Goal seeking behavior

He proposed that environment is
best model and not representation
Homeostasis

These propositions revolutionized
the way of thinking !!
Learning abilities
The Rationale and Objectives
“New Technologies and design approaches for building
physically embodied intelligent agents and artefacts,
with emphasis on the relationship between shape,
function and the physical and social environment”
What is Embodied Intelligence
 Embodied Intelligence (EI) is a mechanism that learns




how to survive in a hostile environment
Mechanism: biological, mechanical or virtual agent with
embodied sensors and actuators
EI acts on environment and perceives its actions
Hostility: direct aggression, pain, anxiety or scarcity of
resources
EI learns so it must have associative self-organizing
memory
Source: Ohio Univ CSE
Continued ...
 It should have a purpose of being
 it should maintain and pursue multiple goals, choosing
which goal to implement based on the environmental
conditions.
 In addition, the complexity of a creature’s behavior
would reflect the complexity of the environment in which
it operates rather than its own.
Source: Ohio Univ CSE
Waste Allocation Load Lifter – Earth Class
 He is better known as Wall-e
 He stays in a very hostile
environment
 reflects the complexity of the
environment it stays in !
 Has an excellent purpose of
being
 Can collect garbage, repair his
own tire and perform a lot more
stuff  multiple goals !!
 In nutshell  excellent example
of embodied intelligence
Source : Pixar Animations
How to create machine intelligence?
Specialized Problems in AI
 Knowledge Representation
 Natural language and scene understanding
 Semantic Cognition
 Reinforcement Learning
Nature Vs Machines

It took nature over 3 billion years to create insects

200 million more years to create mammals

15 million years for the transformation of great apes to modern man about 3
million years ago

Major developments of the civilized world within the last 10,000 years.

It seems that in nature it is easier to append a primitive brain to create a
complex brain.

While this may justify an approach in which a machine’s reflexes are
developed first, the lack of a mechanism to add complexity at a low design
cost is a major problem that cannot be left to chance.
Challenges of Embodied Intelligence
Outline
 Traditional Artificial Intelligence
 Embodied Intelligence (EI)
 Challenges of EI
 We need to know how
 We need means to implement it
 We need resources to build and sustain its operation
 Promises of EI
 To economy
 To society
Intelligence
Evolve
Solve
Plan
Comprehend
Reason
Learn
Think
Traditional AI
 Attempt to simulate “highest”
human faculties like language,
reasoning, problem solving
Embodied Intelligence
 knowledge is implicit in the fact
that we have a body
 Embodiment is a foundation for

Brain is taken to be an abstract
problem solver
brain development
 Intelligence develops through
 Environment model based
approach. Note that
environment is extremely
difficult to model.
constant interaction with
environment
 Problems are identified and
solved by goal seeking behavior
 Pre-specified problems are
solved using abstract ways
Source : Ohio Univ
Design principles of intelligent systems
Interaction with complex environment
cheap design
ecological balance
redundancy principle
asynchronous
parallel, loosely coupled processes
sensory-motor coordination
value principle
Source : Design Principles for Intelligent Systems
Department of Information Technology, University of Zurich
Agent
Embodiment of Mind
 Necessary for development of
intelligence
 Hosts brain’s interfaces that
interact with environment
 Not necessarily constant or in
the form of a physical body
Embodiment
Intelligence
core
 Boundary transforms modifying
brain’s self-determination
Environment
Continued ...
 Brain learns own body’s dynamic
 Self-awareness results from
identification with own
embodiment
 Embodiment can be extended
by using tools and machines
 Successful operation depends on
correct perception of
environment and own
embodiment
The Brain
 While we learn it’s functions can
we emulate it’s operation ?
How can we design intelligence?
 We need to know how
 We need means to implement it
 We need resources to build and sustain its
operation
Name: Dav
Source: MSU Univ CSE
Requirements for Embodied Intelligence
 State oriented

Learns spatio-temporal patterns

Situated in time and space
 Learning

Perpetual learning

Screening for novelty
 Value driven
 Goal creation

Competing goals
 Emergence

Artificial evolution
 Self-organization
EI Interaction with Environment
Agent Architecture
Reason
Short-term Memory
Perceive
Act
RETRIEVAL
LEARNING
Long-term Memory
INPUT
OUTPUT
Task
Environment
Simulation or
Real-World System
We need to develop ..
 Sensory Interfaces




Active Vision
Speech Processing
Tactile, Smell, Taste, Temperature, Pressure Sensing
Additional Sensing
 Infrared, Radar, Ultrasound, GPS, Etc.
 Can Too Many Senses Be Less Useful?
 Reinforcement Interfaces
 Energy, Temperature, Pressure, Acceleration Level
 Teacher Input
 Motor Interfaces
 Arms, Legs, Fingers, Eye Movement
Continued ...
 Algorithmic Solutions For
 Association, Memory, Sequence Learning, Invariance Building,
Representation, Anticipation, Value Learning (Pain Reduction),
Goal Creation, Planning
 Circuits For Neural Computing
 Determine Organization Of Artificial Minicolumn
 Self-organized Hierarchy Of Minicolumns For Sensing And Motor
Control
 Self-organization Of Goal Creation Pathway
Source : Univ. of Sussex Alastair Channon
Goal Driven Behavior
 Goal driven behavior is one of the




required elements of intelligence
Perceptions and actions are activated
selectively to serve the machine’s
objectives
In the existing EI models, the goal is
defined by designers and is given to
the learning agent
Humans and animals create their own
goals
The goal creation may be one of the
most important elements of EI
mechanism
Source: Janusz A. Starzyk -- Challenges of EI
Goal Creation
 Goals must be built and understood in a
similar way to building perceptions
 complex goals can be understood only if
representations are build
 It should result from EI interaction with
environment, by perceiving successes or
failures of its actions
 essential for developing intelligence
 We will create goals based on simple
structures interacting with sensory and
motor pathways
Source: Janusz A. Starzyk -- Challenges of EI
How can we design intelligence?

We need to know how

We need means to implement it

We need resources to build and
sustain its operation
Doubling (or Halving) times
 Clock speed
2.7 years
 Dynamic RAM Memory “Half Pitch” Feature Size
5.4 years
 Dynamic RAM Memory (bits per dollar)
1.5 years
 Average Transistor Price
1.6 years
 Microprocessor Cost per Transistor Cycle
1.1 years
 Total Bits Shipped
1.1 years
 Processor Performance in MIPS
1.8 years
 Transistors in Intel Microprocessors
2.0 years
 Microprocessor Clock Speed
2.7 years
Source: Chemheritage.org, IEEE Explore IITB
Is It Possible ?
 the area occupied by the new logic must be gradually reduced from
over 60% in 1999 to less than 5% in 2010.
 It is a way to shorten the design time, but it doesn’t create high-value
designs !!!
 Yet the structure of interconnections in human brain is very complex.
 Thus a design of EI could be tremendously costly even if we know how
to build it.
Promises of embodied intelligence
 To society
 Advanced use of technology
 Robots
 Tutors
 Intelligent gadgets
 Society of minds
 Superhuman intelligence
 Progress in science
 Solution to societies’ ills
 To industry
 Technological development
 New markets
 Economical growth
Name : Sail
Source: MSU Univ CSE
Sounds like science fiction
 If you’re trying to look far
ahead, and what you see
seems like science fiction, it
might be wrong.
 But if it doesn’t seem like
science fiction, it’s definitely
wrong.
Name : Wall-E
Source: Pixar Animations
THANK YOU
References









Motivation in Embodied Intelligence (2005)– janusz Starzyk (Ohio Univ.
USA)
Challenges of Embodied Intelligence (2001) - Janusz A. Starzyk, Yinyin
Liu, and Haibo He (Ohio Univ. USA)
Moravec, H.P. (1999) -- Rise of the Robots
Pixar Animations  http://www.pixar.com/featurefilms/walle/
The Evolutionary Emergence route to Artificial Intelligence -- Alastair
Channon (UNIVERSITY OF SUSSEX 1995-96)
Design Principles for Intelligent Systems (2003)  Rolf Pfeifer1, Fumiya
Iida1, Josh Bongard2 Department of Information Technology, University
of Zurich
Michigan State Univ. http://www.cse.msu.edu/ei/
Chemheritage.org & IEEE Explore IITB
Wikipedia